IBM Watson Could Boost Your Medical Device

Chris Newmarker

June 14, 2016

3 Min Read
MDDI logo in a gray background | MDDI

Developing a smartphone or tablet app related to a medical device? IBM is introducing Watson's patient record deciphering know-how services, as well as natural language processing, as a value add, according to the principal investigator for the Watson EMR Analyzer (EMRA). 

Chris Newmarker

Picture a smart tablet app that allows radiologists to quickly view images. Now what if the tablet app also listed the patient's medical problems on the left, with an option to further explore the clinical notes, lab results, medications, etc. related to each medical problem?

That's an app idea IBM is actually working on with its Watson supercomputer, Murthy Devarakonda, an IBM research scientist who is principal investigator for the Watson EMR Analyzer (EMRA), said Tuesday at MD&M East in New York City.

EMRA, first introduced in 2013, is able to sort through a patient's EMR, which can be up to 20 MB in size, and provide "quick and accurate insights." EMRA is able to generate a problem list and patient record summary, and also provide a semantic "find" search for a health provider.  

Such insights could prove crucial because it is hard these days for health providers to keep track of important data about a patient's health, Devarakonda said. 

"When you lived in a small town, a physician knew all about you. But these days, it's more transactional. ... They may have never seen you. They may have never seen you in a while. How can we meet the information needs of a physician when they are treating a patient?" Devarakonda said.

IBM is presently working on adding natural language Q&A on EMR content and even advanced clinical insights to EMRA. 

The key was to get Watson to sort through the reams of clinical notes plain text data, as well as tables of medications, lab results, and more.

To overcome this challenge, IBM developed a Unified Medical Language System (UMLS). "Each word is mapped into an ID, and there are synonyms for it," Devarakonda said.

The relationship between various terms is based on ontology from the National Library of Medicine. Devarakonda showed an example of clinical notes in which orange highlighted a "disease or syndrome" such as diabetes mellitus, while purple highlighted a "sign or symptom" such as "erythema."

Watson's machine learning them comes into play by connecting the dots between all of the different pieces of data in the patient's record, and providing a score to various health problems to determine which are real problems and which are diseases that may have been mentioned in passing (or related to relatives) but not a problem for the patient. 

"This connecting and having this problem list creates a nice abstracted summary that physicians can use to get them up to speed. ... The model takes all these features and then creates a score for each of the problems. The potential problems that score high are real problems for the patients," Devarakonda said.

IBM's experiments, according to Devarakonda, have demonstrated that Watson is able to demonstrate an accurate problem list, too. 

Chris Newmarker is senior editor of Qmed and MPMN. Follow him on Twitter at @newmarker.

Like what you're reading? Subscribe to our daily e-newsletter.

Sign up for the QMED & MD+DI Daily newsletter.

You May Also Like